José Caballero

Timeline

1. April 2014

   Dictionary learning for cardiac MRI

   In April 2014, he coauthors Dictionary Learning and Time Sparsity for Dynamic MR Data Reconstruction. The paper presents an iterative algorithm that applies dictionary learning to reconstruct cardiac cine data with Cartesian undersampling.

2. September 2015

   A PhD in fast MRI

   Before his doctorate, he completes an MEng in Electrical and Electronic Engineering and an MSc in Communications and Signal Processing at Imperial College London. In September 2015, he earns a PhD in Computing at Imperial after doctoral studies in the Biomedical Image Analysis (BioMedIA) group under Daniel Rueckert and Jo V. Hajnal. His thesis, Dictionaries for fast and informative dynamic MRI acquisition, introduces adaptive spatio-temporal dictionaries for cardiac cine MRI reconstruction and joint image analysis.

3. 2015

   Magic Pony and Twitter

   He joins Magic Pony Technology in 2015, where he designs and develops real-time super-resolution algorithms for mobile devices. After Twitter acquires Magic Pony Technology, he works at Twitter on video compression, enhancement and machine-learning solutions to process media content.

4. 2016

   The sub-pixel CNN

   In 2016, he coauthors a CVPR paper presenting a convolutional neural network that performs 1080p video super-resolution in real time on a single K2 GPU. The network extracts feature maps in low-resolution space and uses a learned sub-pixel convolution layer to create high-resolution outputs.

5. 2017

   SRGAN and spatio-temporal super-resolution

   His 2017 CVPR paper introduces spatio-temporal sub-pixel convolution networks that exploit temporal redundancies across consecutive frames, along with a trainable motion-compensation method that improves video super-resolution accuracy at real-time speed. In 2017, he also coauthors the CVPR paper presenting SRGAN, a generative adversarial network for photo-realistic single-image super-resolution at fourfold upscaling factors.

6. April 2017–December 2017

   Deep neural networks for cardiac MRI

   In April 2017, he coauthors A Deep Cascade of Convolutional Neural Networks for Dynamic MR Image Reconstruction, which reconstructs dynamic 2D cardiac MRI sequences from undersampled data to accelerate acquisition. In December 2017, he coauthors Convolutional Recurrent Neural Networks for Dynamic MR Image Reconstruction, proposing a convolutional recurrent neural network architecture to reconstruct cardiac MRI images from highly undersampled k-space data.

7. July 2019–August 2020

   Emoji embeddings for visual sentiment

   In July 2019, he coauthors Smile, Be Happy :) Emoji Embedding for Visual Sentiment Analysis, training a deep neural image embedding to predict emojis across a dataset of four million Twitter images. In an August 2020 revision, the authors show that the emoji embedding transfers across domains to visual sentiment analysis and fine-grained emotion recognition.

8. As of October 2026

   Lead research engineer at Epic

   As of October 2026, he is a lead research engineer at Epic Games.